JPH05314094A - Sales prediction supporting device - Google Patents

Abstract

PURPOSE:To provide the supporting device which facilitates the operations, reduces the generation of error as well and is convenient for users to finally judge prediction by simplifying a transient probability distribution function or a membership function into the form of practical use. CONSTITUTION:Sales result data in the past inputted from an input device are stored in a result storage file part 1. Based on these data, a trend component arithmetical prediction part 2 performs the extraction and prediction of trend components. The trend component is the tendency of sale and is smoothed by removing the increase of sales results under any temporary and special conditions, and a predictive value is calculated. Based on this calculated total predictive value, the transient probability distribution function is triangularly expressed for each item of products and displayed at an output device 10. The bottom side of the triangle is defined as the size of fluctuation width, the apex is defined as the total predictive value for each item and when the height is fixed for all the items, the much smoother triangle is judged as a product for which the sales fluctuation in the past is big and the more sale can be estimated corresponding to following sales strategy.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is used as a support device for predicting the sales of goods in order to make a production plan for a factory or a sales plan for goods. The present invention controls the inventory quantity of products to the minimum, eliminates missing items and delays in delivery of product orders, and further promotes sales activities rationally. Used as a support device for making sales forecasts. The device of the present invention utilizes a computer device, and all calculations and displays are executed under program control.

[0002]

2. Description of the Related Art It is important to minimize the inventory quantity of products and reduce the inventory cost in order to keep the cost of the products small. If you do, you will lose credit for the product and you will have to pay a penalty depending on the contract.
Also, when promoting sales of products through sales activities, where to set the next sales target is an important factor.

For this reason, conventionally, the sales record of the past relevant merchandise, which is totaled every month, is recorded in a computer for a long time, which is displayed in a graph on the display screen of the computer, and the administrator in charge while watching this display. It is common practice to forecast sales volume for the next month. In this case, for example, if the product is a detergent, the detergent has a seasonal variation with a large demand in the summer and a small demand in the winter. Is considered as a trend component. Therefore, a technique is known in which a trend component is extracted by a computer, and the extracted trend component is displayed in a graph displayed on a display screen.

On the other hand, OR (Operation Research)
In the research field of, various algorithms have been known so far for sales forecasting or production management. There are some known reports in which a computer device is used to perform calculations with a large number of data using this algorithm.

[0005]

Even if the sales forecast is executed by using a high-grade algorithm or a high-grade device, the actual sales are not only influenced by the laws of nature but also fluctuated by human factors. Therefore, the calculation result cannot be used as it is for business policy. Eventually, it will be a human decision as the decision of the business or management of the company. High-end algorithms and high-end devices are the research results for making accurate forecasts as much as possible, but for sales forecasts that are used on a daily basis, the types and amounts of data to be input are large and the operations are complicated. Therefore, it is not always a convenient support device for the user to use and make a judgment such as an error in using the device.

The present invention focuses on the fact that a transition probability distribution function or a membership function, which performs an operation based on a transition process of fluctuations, is a useful logic for sales forecasting or production management, and this is put to practical use. Simple, simple to use, the type and amount of data to be input is small, and the operation is easy.
It is an object of the present invention to provide a support device which is small in the occurrence of errors and is convenient for people to finally make a prediction judgment.

[0007]

According to the present invention, a sales record file in which sales records of one or a plurality of products are accumulated, and the sales record accumulated in the recording record file are taken out, and the sales records are periodically reproduced. Component calculation means for extracting a trend component that is a large variation, a fluctuation component calculation means for removing the trend component from the sales record to extract a variation component, and a prediction of this variation component for the next coming period. Fluctuating component predicting means, and trend component predicting means for predicting the trend component by extrapolating the trend component for the next coming period,
And a prediction calculation unit that sets the sum of the trend component prediction value predicted by the trend component prediction unit and the fluctuation component prediction value predicted by the fluctuation component prediction unit as a comprehensive prediction value for the next coming period. In the sales forecast support apparatus, the fluctuation component predicting unit divides the performance value extracted from the performance storage file into a plurality of clusters, and the fluctuation component H (k) of the latest performance value is in the i-th cluster i The actual value fluctuated from the second cluster is Δaj
(J = 1, 2, ..., N), the average value is

[0008]

[Equation 4] And a means for calculating the fluctuation component predicted value as H ′ (k + 1) = H (k) + Δi.

Further, the actual value, the trend component, and the fluctuation component are displayed on a time axis, and the amplitude of the fluctuation component extracted by the fluctuation component computing means is set as a vertex position and the base length is set as the base length. It is desirable to provide means for displaying a triangle whose height is the length of the next period. Furthermore, the fluctuation component H of the latest actual value
When (k) is in a cluster in which the fluctuation component is zero, it is desirable that the fluctuation component prediction value be zero and the trend component prediction value be the comprehensive prediction value.

As another aspect of the present invention, a history storage file in which sales records of one or a plurality of products are stored, and trend components associated with seasonal fluctuations are extracted from the sales records stored in the sales record file. Trend component calculating means, a fluctuation component calculating means for removing the trend component from the sales record to extract a fluctuation component, a fluctuation component predicting means for predicting the fluctuation component for the next coming period, and Trend component predicting means for predicting the trend component by extrapolating the trend component for the period of arrival, and the trend component prediction value predicted by the trend component predicting means and the fluctuation predicted by the fluctuation component predicting means. Sales including a prediction calculation means for making the sum of the component predicted values into a total predicted value for the next coming period In the prediction support apparatus, the fluctuation component prediction means divides the performance value extracted from the performance storage file into a plurality of clusters, and the fluctuation component H (k) of the latest performance value is in the i-th cluster, and the fluctuation thereof Ingredient H
When (k) has an increasing tendency with respect to the fluctuation component actual value H (k−1) of the previous period, the actual value increased and varied from the i-th cluster is Δcj, (j = 1, 2, ..., N). ) As
The average value of the actual values that increased and changed

[0011]

[Equation 5] When the fluctuation component H (k) has a decreasing tendency with respect to the fluctuation component actual value H (k−1) of the previous period, the actual decreasing value of the i-th cluster is set as ej, and the decreasing fluctuation The average value of the actual results

[0012]

[Equation 6] And a means for calculating the fluctuation component predicted value as H ′ (k + 1) = H (k) + Δi.

The actual value, the trend component, and the fluctuation component are displayed on the time axis, and the amplitude of the fluctuation component extracted by the fluctuation component computing means is the base length with the total predicted value as the vertex position. It is desirable to provide means for displaying a triangle whose height is the length of the next period. When the fluctuation component H (k) of the latest actual value is in the cluster in which the fluctuation component is zero, it is preferable to set H ′ (k + 1) = 0 and use the trend component prediction value as the comprehensive prediction value. Further, it is desirable to include means for calculating the total predicted value, the fluctuation component predicted value, and the triangle for a plurality of product items.

[0014]

[Operation] Past sales record data input from the input device (keyboard or the like) is accumulated in the record storage file section. Based on this data, the trend component calculation / prediction unit extracts and predicts the trend component. The trend component is the tendency of sales, which is smoothed by removing the growth of the sales performance under a temporary and unique condition to calculate the forecast value. This is to prevent peculiar data from affecting the accuracy of the predicted value.

Next, the fluctuation component calculation / prediction unit extracts and predicts the fluctuation component. The fluctuation component is the fluctuation in sales and the fluctuation range. The fluctuation range is decomposed into clusters of several stages, and the predicted value of fluctuation is calculated for each cluster, but it is further classified according to the increasing trend and the decreasing trend, and the increasing trend of the past increasing trend. The predicted value is calculated based on the data, and when the tendency is decreasing, the predicted value is calculated based on the data of the past decreasing tendency.

Based on the predicted values of the trend component and the fluctuation component calculated in this way, the comprehensive prediction calculation unit calculates the sales predicted value for the next month. The result is output from the comprehensive predicted value data output interface.

Based on the calculated total predicted value, the transition probability distribution function for each product item can be represented by a triangle and displayed on an output device (CRT or the like).
If the bottom of the triangle is represented by the size of the fluctuation range, the apex is represented by the total forecast value for each item, and the height is constant for all items, the smoother the triangle, the greater the fluctuation in sales in the past and the future sales. Depending on the strategy, it is judged that there is still a possibility of selling.

[0018]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The configuration of the device of the embodiment of the present invention will be described with reference to FIG. FIG. 1 is a block diagram of an apparatus according to an embodiment of the present invention.

The present invention takes out the sales record file unit 1 in which the sales records of one or a plurality of products are accumulated for a sufficiently long period of time, and the sales records accumulated in the record storage unit 1 are taken out and sold. A trend component calculation that extracts a trend component associated with periodic fluctuations and seasonal variations from the actual results, removes the trend component from the sales results, extracts the fluctuation component, and predicts the fluctuation component for the next coming period. A prediction unit 3, a trend component calculation prediction unit 2 that predicts a trend component by extrapolating the trend component for the next coming period, and a trend component prediction value predicted by the trend component calculation prediction unit 2. And the variation component prediction value predicted by the variation component calculation prediction unit 3 as a total prediction value for the next coming period. In the sales prediction support device including the prediction calculation unit 4, the fluctuation component calculation prediction unit 3 divides the long-term result values extracted from the result storage file unit 1 into a plurality of clusters, and changes the latest result value fluctuation component. H
(K) is in the i-th cluster, and the actual value that has changed from the i-th cluster in an increasing tendency is Δaj, (j = 1, 2,
, N), when the fluctuation component H (k) tends to increase with respect to the fluctuation component actual value H (k−1) one period before, the average value is calculated.

[0020]

[Equation 7] Then, when the fluctuation component H (k) has a decreasing tendency with respect to the fluctuation component actual value H (k−1) one period before, the actual value decreased from the i-th cluster is changed to ej (j = 1,
2, ..., N), the average value is

[0021]

[Equation 8] And a means for calculating the fluctuation component predicted value as H ′ (k + 1) = H (k) + Δi.

The actual value, the trend component, and the fluctuation component are displayed on the time axis, and the amplitude of the fluctuation component extracted by the fluctuation component calculation / prediction unit 3 is set at the apex position. The CRT 10 is provided with a means for displaying a triangle with the length as the length and the length of the next period as the height. Further, when the fluctuation component H (k) of the latest actual value is in the cluster in which the fluctuation component is zero, H ′ (k + 1) = 0 is set and the trend component prediction value is set as the comprehensive prediction value. Further, it is configured to include means for calculating the comprehensive predicted value, the fluctuation component predicted value, and the triangle for a plurality of product items.

Next, the operation of the apparatus of the present invention will be described with reference to FIG. FIG. 2 is a flow chart showing the operation of the apparatus of the present invention. Enter the sales results over a sufficiently long period and first make a trend forecast. The trend component calculation algorithm consists of model fitting consisting of three steps and model selection by AIC (Akaike's information criterion). The model is A from the following four formulas
The one with the smallest IC value is selected, model 1: y = a + bk + csin (πk / 6) +
d · cos (πk / 6) model 2: y = a + b · sin (πk / 6) + c · co
s (πk / 6) Model 3: y = a + b · k Model 4: y = a (however, a, b, c, d: parameter, k: total month) Model 1 is a seasonal fluctuation (a sine wave with a one-year cycle). ) And a linear (linear) trend up or down trend is considered. The model 2 is a model that considers only seasonal fluctuations, and the model 3 is a model that considers only rising or falling of a linear trend. Further, the model 4 is a model in which there is an almost constant sales volume. Since the model is mechanically selected for each month, the model selected for each month may differ depending on the data set (item).
However, for products for which changes in sales volume other than peculiar data can be expressed by any of these four models, changes in the selected model have a significant impact on the forecast of trend components. Not considered.
Model fitting is done in three steps. The first step uses a 3-month mobile median, and the second step is the usual least squares method (OLS: Or).
The model fitting is carried out by means of the directory least squares. The third step is the iterative weighted least squares method (IWLS: Itera).
ted Weighted LeastSquare
The final parameter calculation is performed according to s).

Next, the first step will be described.
The main purpose of the first step is peak removal. "3
"Monthly mobile median" is a method that uses the median (median) of each month and the data for three months including the month before and after that. A 3-month mobile median will be described with reference to FIGS. 3 and 4. 3 and 4 are 3
It is a figure explaining a mobile median for six months. In FIG. 3, in the case of July 1989, the actual results are in the order of August 1989 <July 1989 <June 1989, so the mobile median value is the July data value itself. In the case of December 1989, the actual results are January 1990 <November 1989 <December 1989, so the mobile median value becomes the data value for November. With a three-month mobile median, large fluctuations less than once every three months can be eliminated. The characteristic of the mobile median is that, as shown in FIG. 4, when the data value is monotonically increasing or decreasing, the original data values are the same except for the data at both ends. 9 in FIG.
From January to July 0, the data of the mobile median and the actual data are almost the same. In addition, when the mobile median is used, dots at both ends (in Fig. 3 and Fig. 4, 89
Since the values for month and August 1990) are not defined, the number of data will be two less than the original number of data.

Next, the second step will be described with reference to FIG. FIG. 5 is a diagram for explaining the least squares method.
The purpose of the second step is to give an initial value for the third step. Assuming that the original number of data is K, K-2 data excluding two months at both ends are obtained by the moving median in the first step. A model is fitted to these K−2 pieces of data by regression using the usual least squares method. Since the fitting is performed for each of the four models shown at the beginning, the least squares calculation is performed four times. The least squares method is the residual 2
Is the minimization of the sum of sums,

[0026]

[Equation 9] (However, r: residual), and a regression as shown in FIG. 5 is obtained.

Next, the third step will be described with reference to FIG. FIG. 6 is a diagram for explaining the iterative weighted least squares method. Although the mobile median can remove the peak once every three months, it has little effect on the unique data that continues for two months. Taking the sales record of Figure 6 as an example, 1
The sales volume in January and December is particularly large compared to other months. Therefore, it can be seen that the regression equation (initial solution) obtained in the second step is greatly affected by those two-month data. In order to remove this effect, the iterative weighted least squares method is used. Repeated weighted minimum 2
The multiplicative method is a method of minimizing the sum of squares of residuals by using the ordinary least square method.

[0028]

[Equation 10] (However, r: residual) based on the other hand, (normalized) minimize the sum of appropriate even function ρ of residuals

[0029]

[Equation 11] (However, u: standardized residual u _k = r _k / σ, σ: estimated value of standard deviation). For convenience, this formula

[0030]

[Equation 12] (However, w: weight function) can be rewritten as a weighted least squares method. The time [rho is called Biweight (or bisquare), ρ (u) = (B 2/6) (1- (1- (u / B) 2) 3) if | u | <B or, [rho ^{(u) = (B 2/} 6) otherwise ( except B: constant (6.0) in the present invention embodiment) is used, in which case the weight function w is from the derivative [rho 'of [rho w (u) = ρ ′ (u) / u = (1- (u / B) ² ) ² if | u | <B or ρ ′ (u) / u = 0 otherwise. In this case, the constant B is set to 6.0, so
For data that is more than 6 times the standard deviation from the regression line, the weight is 0, which means that the data is substantially removed. A standardized residual is required to determine the weight, and a standard deviation must be estimated to obtain the standardized residual. If the standard deviation is estimated as an initial value (regression with respect to the moving median value),

[0031]

[Equation 13] (However, r ': difference between the value of the moving median and the value of the regression equation,
K: number of data (months)) is used. Since the second-step least-squares method is performed on the moving median value, not the original data, the estimated standard deviation is expediently doubled from the usual case. After the regression equation is determined,

[0032]

[Equation 14] Is used as an estimate of standard deviation. The estimated value of the standard deviation changes as the weight of each data changes. On the contrary, the weight of each data changes if the estimated value of the standard deviation changes.
Therefore, iterative calculation is performed until these changes are sufficiently small. In this case, it means that the change of the weight of the data is small enough as the convergence judgment condition.

[0033]

[Equation 15] (However, w _k : the weight used to obtain the regression equation,
w ′ _k : weight calculated from the obtained regression formula) is used.

Next, the change in the weight of the sales record of FIG. 6 is shown in FIG.
Will be described. FIG. 7 is a diagram illustrating a change in weight. FIG. 7 shows changes in weighting when the model 1 is applied to FIG. 7 (a)-(d),
Trend components (that is, lines with weight 1) and weights 0.
The contour lines at 9, 0.5, 0 are shown. From FIG. 7A, it can be seen that the initial weight obtained from the regression equation for the moving median value does not have a weight of 0 in December. Further, the weight of the data in November is 0.9 or more. The result of the weighted least squares method using the “initial weight” as a weight is the trend component of “iteration 1” in FIG. 7B. When the weights are recalculated from this result, it can be seen that the weight of the December data is 0 and the weight of the November data is also low. As the iterative calculation process is followed, the data for November has a weight of 0 obtained from the result of the third iterative calculation in FIG. 7 (d), which indicates that it does not affect subsequent calculations. Recognize. In this case, the convergence condition is satisfied in the fifth iteration of FIG. 7E, and the iterative calculation is completed. Compared with the case where the regression up to the second step was performed on the moving median value, November, 1
It can be seen that the influence of unique data such as February is sufficiently small. Just as the second step performed calculations for each model, the third step also performed calculations for each model. Of course, the initial solution for each model uses the regression equation calculated in the second step for each model.

Next, model selection will be described. AIC (Akaike's information criterion) is used for the problem of which model is selected from the four fitted models. In this case, the AIC

[0036]

[Equation 16] (Where p is the number of parameters of the model), it is represented by an equation of 4 in the case of model 1, 3 in the case of model 2, 2 in the case of model 3, and 1 in the case of model 4. The value of the above equation is calculated for each convergence value of the four models, and the model having the smallest value is selected. The regression equation of the model selected by the result of this selection is used as the trend component. Further, a value obtained by substituting K + 1 into the selected regression equation is calculated as the expected value for the next month.

Next, the procedure for calculating the fluctuation component using this trend component and calculating the fluctuation component predicted value will be described with reference to FIG. FIG. 8 is a diagram for explaining the calculation of the fluctuation component predicted value. FIG. 8A shows the relationship between the trend component T and the sales result data (T ′ (k + 1): the predicted variation component of the trend component T). The fluctuation component H is calculated using the relationship of fluctuation component H = (sales performance data) − (trend component T), and a graph as shown in FIG. 8B is obtained. Further, the means for calculating the variation component prediction divides the variation component data into nine clusters at equal intervals as shown in FIG. 8C, and next to the past variation component data in the cluster having the variation component data of this month. Calculate the average of fluctuations over the month. Then, the value is added to the fluctuation component data of this month to be the predicted value of the fluctuation component of the next month. In the case of FIG. 8C, since the fluctuation component H (k) of this month belongs to the cluster “8”, the fluctuation component prediction value H is calculated from the fluctuation component H (k) of this month.
The fluctuation amount Δ8 to (k + 1) is the average value of the fluctuation amounts (Δa, Δb) of the fluctuation components belonging to the cluster “8” in the past, and Δ8 = (Δa + Δb) / 2. Therefore, the fluctuation component predicted value H ′ (k + 1) is H ′ (k + 1) = H (k) + Δ8. The values of averages Δ1 to Δ9 of the fluctuation amount are determined for each cluster, but when the fluctuation component of this month is in the vicinity of 0, the prediction result of the fluctuation component conversely deteriorates the prediction result. Therefore, the cluster having the fluctuation component 0 and the clusters in the vicinity thereof, “6” in FIG.
If there are fluctuation components of this month in the clusters “5” and “7” in the vicinity of and, the fluctuation component prediction value is set to 0, and the sales prediction value of the next month = trend component prediction value. The fluctuation amount Δi of the cluster i is shown in FIG.
In the case shown in (d), Δi = (Δa + Δb + Δc + Δd) / 4 was set, but this gives the same predicted value regardless of whether the sales tend to increase or decrease. Therefore, the prediction values for the case where the fluctuation component of this month in cluster i has an increasing tendency and a decreasing tendency as compared with the previous month will be given separately. That is, when the fluctuation component H (k) of this month has an increasing tendency from the fluctuation component H (k−1) of the previous month, the fluctuation component existing in the cluster i in the increasing tendency is set as the average value of the fluctuation amounts to the next month. To do. Looking at the cluster i in FIG. 8D, there is an increasing tendency from the previous month to this month. Therefore, if the average value Δi ′ of the increasing tendency in the same cluster in the past is obtained, Δi ′ = (Δb + Δd) / 2. .. Therefore, the fluctuation component predicted value is H ′ (k + 1) = H (k) + Δi ′. In this way, the sales forecast value for the next month is given as the sum of the trend component forecast value and the fluctuation component forecast value.

Here, the relationship between the total production amount and the transition probability will be described with reference to FIG. FIG. 9 is a diagram showing variation components and transition probabilities. This shows the transition of the fluctuation component on the left side of FIG.
The transition probability distribution at the time point k is schematically shown on the right side. In this figure, the vertical axis is the displacement of the fluctuation component, and the horizontal axis is the transition probability. The maximum when the probability distributions of the constituent elements m and n (corresponding to the item) are ρ _m and ρ _n , respectively. The displacement amounts of the fluctuation components when the probability is given are shown as X _m and X _n . That is, these X _m and X _n are the predicted values of the fluctuation component itself. Normally, the total production G is not always equal to the sum of the predicted values Xi,

[0039]

[Equation 17] Becomes

Next, the procedure for representing the transition probability distribution function by a triangle will be described with reference to FIG. FIG. 10 is a diagram illustrating a procedure of representing a transition probability distribution function by a triangle. The relationship between the transition probability and the displacement as shown in FIG. 9 is represented by a triangle as shown in FIG. Next, as shown in FIG. 10B, the difference between the total production amount G and the total predicted value is optimally distributed to each component i (iεN).

Next, the procedure for determining the optimum distribution from the transition probability distribution function will be described with reference to FIG. FIG. 11 is a diagram illustrating a procedure for determining the optimum distribution from the transition probability distribution function. Here, by setting ρi (ΔXi) = yi, an increase / decrease in transition probability is expressed by an equation. A linear optimal problem is solved using these transition probability distribution functions so as to achieve a total production amount of 1000 units, provided that if G-ΣXi> 0, then ai = -1 / ΔX _i ma
If x G-ΣXi <0, then ai = -1 / ΔX _i min (where ΔX _i max is the maximum value of the fluctuation component, ΔX _i mi
(n represents the minimum value of the fluctuation component), the transition probability distribution function is expressed as shown in FIG. Becomes Here, as shown in FIG. 11B, assuming that y1 = y2 = ... = yn = α (0 <α ≦ 1), ΔX _i satisfying these equations is set within the constraint of T−ΣX _i = ΣΔX _i. Just ask.

As described above, a triangle having the maximum and minimum values of the fluctuation component as the bases and the fluctuation component predicted value calculated by the process shown in FIG. This triangle approximates the transition probability distribution function that gives the transition probability ρ, and determines this for each component (corresponding to the item in this embodiment). Based on the transition probability distribution function determined in this way, the difference between the total of the predicted values and the sales target amount is distributed to the constituent elements as shown in FIG. In this case, the transition probability distribution function is a function that represents sales sensitivity. By defining the transition probability distribution function as a triangle,
Not only can it be made in a form that is easy for anyone to understand, that is, a form that is advantageous for data maintenance, but this also makes it possible to speed up computation.

Next, the transition probability distribution function represented by a triangle with respect to the fluctuation component data of the four products will be described with reference to FIG. FIG. 12 is a diagram showing the relationship between the variation component data of four products and the triangle of the transition probability distribution function. As shown in FIG. 12, a transition probability distribution function represented by a triangle is obtained from the variation component data. When comparing the triangles of item 1, 2, 4 and item 3, the triangle of item 3 has a long base, It can be seen that the amplitude of fluctuation is large. That is, it can be inferred that there is still a considerable possibility that the sales will exceed the predicted value depending on the sales strategy. Therefore, it can be estimated that the difference between the predicted value and the sales target is assigned to item 3 at a large rate. On the other hand, if for some reason (for example, productivity improvement, new market development, trends of other companies, etc.), the transition probability distribution does not fit the future situation, a means for responding to it is necessary. FIG. 12A is an image when the fluctuation component and the transition probability distribution function of each item are displayed on the device of the present invention. FIG. 12B shows a diagram obtained by rotating the transition probability distribution function by 90 degrees. Here, by directly changing the transition probability distribution function (peak position and displacement width), it becomes possible to use the above-mentioned coping means. This function is also a simulation function for optimization, and is also effective for forecasting series (new products, seasonal products) with a small amount of data. In this way, since the information of the transition process is included, it becomes possible to optimally distribute the components to various characteristics. FIG. 13 is a diagram showing a sales schedule forecast by the new method. As a result, as shown in FIG. 13, the forecast accuracy of the sales schedule is improved, and the man-hour until the sales schedule is formulated can be shortened.

[0044]

As described above, according to the present invention, the transition probability distribution function or the membership function, which is operated based on the transition process of fluctuation, is a useful logic for sales forecasting or production management. Focusing on this, simplifying this to a practical form, the type and amount of data to be input is small, the operation is easy, the occurrence of errors is small, and it is convenient for people to finally make prediction judgments. .. The transition probability distribution function is simplified to a practical form, the type and amount of data to be input is small, the operation is easy, and the number of errors is small,
Based on the sales forecast of components having various characteristics, it becomes possible to optimally distribute the target value for each item. As a result, the forecast accuracy of the sales schedule is improved, and the man-hours until the formulation of the sales schedule can be shortened.

[Brief description of drawings]

FIG. 1 is a block diagram of an apparatus according to an embodiment of the present invention.

FIG. 2 is a flowchart showing the operation of the apparatus according to the present invention.

FIG. 3 is a diagram illustrating a mobile median.

FIG. 4 is a diagram illustrating a mobile median.

FIG. 5 is a diagram illustrating a least squares method.

FIG. 6 is a diagram illustrating an iterative weighted least squares method.

FIG. 7 is a diagram for explaining changes in weighting.

FIG. 8 is a diagram illustrating calculation of a fluctuation component predicted value.

FIG. 9 is a diagram showing fluctuation components and transition probabilities.

FIG. 10 is a diagram illustrating a procedure of representing a transition probability distribution function by a triangle.

FIG. 11 is a diagram illustrating a procedure for determining an optimum distribution from a transition probability distribution function.

FIG. 12 is a diagram showing a relationship between variation component data of four products and a triangle of a transition probability distribution function.

FIG. 13 is a diagram showing a forecast of sales schedule by a new method.

[Explanation of symbols]

 1 Actual result storage file part 2 Trend component calculation prediction part 3 Fluctuation component calculation prediction part 4 Total prediction calculation part 10 CRT

Claims (7)

[Claims] 1. A result storage file in which sales results of one or a plurality of products are accumulated, and the sales results accumulated in the result storage file are taken out, and a trend component that is a periodic fluctuation is extracted from the sales results. A trend component calculating means, a fluctuation component calculating means for removing the trend component from the sales record to extract a fluctuation component, a fluctuation component predicting means for predicting the fluctuation component for the next coming period, Trend component predicting means for predicting the trend component by extrapolating the trend component for the period of arrival, and the trend component predicted value predicted by the trend component predicting means and the fluctuation predicted by the fluctuation component predicting means. Sales including a prediction calculation means for making the sum of the component predicted values into a total predicted value for the next coming period In predicting support device, the fluctuation component predicting means, dividing the actual value extracted from the actual storage file into a plurality of clusters, the variation component of the latest actual value H (k) is i
When the actual value that is in the i-th cluster and fluctuates from the i-th cluster is Δaj (j = 1, 2, ..., N), the average value is And a means for calculating the fluctuation component predicted value as H '(k + 1) = H (k) + Δi. 2. The actual value, the trend component, and the fluctuation component are displayed on a time axis, and the amplitude of the fluctuation component extracted by the fluctuation component computing means is the base length with the comprehensive predicted value as a vertex position. 2. The sales forecasting support device according to claim 1, further comprising means for displaying a triangle having a height of the next coming period as a height. 3. The fluctuation component prediction value is set to zero and the trend component prediction value is set to the total prediction value when the fluctuation component H (k) of the latest actual value is in a cluster in which the fluctuation component is zero. 1. The sales forecast support device described in 1. 4. A record storage file in which sales records of one or a plurality of products are stored, and trend component calculation means for extracting trend components associated with seasonal fluctuations from the sales records stored in the record storage file. A fluctuation component calculation means for removing the trend component from the sales record to extract a fluctuation component, a fluctuation component prediction means for predicting the fluctuation component for the next coming period, and a trend for the next coming period. The trend component predicting means for predicting the trend component by extrapolating the component, and the sum of the trend component predicting value predicted by the trend component predicting means and the fluctuation component predicting value predicted by the fluctuation component predicting means. In a sales forecast support device including a forecast calculation unit that provides a comprehensive forecast value for the next coming period, The variation component predicting unit divides the performance value extracted from the performance storage file into a plurality of clusters, and the fluctuation component H (k) of the latest performance value is i.
In the second cluster, whose variation component H (k) is 1
When the fluctuation component actual value H (k-1) before the period is on the increase, the actual value increased from the i-th cluster is changed by Δcj (j
= 1, 2, ..., N), the average value of the increased and changed actual values is When the fluctuation component H (k) has a decreasing tendency with respect to the fluctuation component actual value H (k−1) of the preceding period, the actual value decreased from the i-th cluster is Δej (j).
= 1, 2, ..., N), the average value of the reduced and changed actual values is given by And a means for calculating the fluctuation component predicted value as H '(k + 1) = H (k) + Δi. 5. The actual value, the trend component, and the fluctuation component are displayed on a time axis, and the amplitude of the fluctuation component extracted by the fluctuation component calculation means is the base length with the comprehensive predicted value as the vertex position. 5. The sales forecasting support device according to claim 4, further comprising means for displaying a triangle whose height is the length of the next coming period. 6. When the fluctuation component H (k) of the latest actual value is in a cluster in which the fluctuation component is zero, H ′ (k + 1) = 0 is set, and the trend component prediction value is set as the comprehensive prediction value. Item 4. A sales forecast support device according to item 4. 7. The sales forecast support apparatus according to claim 1, further comprising means for calculating the total forecast value, the fluctuation component forecast value, and the triangle for a plurality of product items.